Microsoft Purview AI Hub setup for Australian tenants — the working walkthrough

Microsoft Purview's AI Hub is now generally available, and at the same time has been renamed to Microsoft Purview Data Security Posture Management for AI — DSPM for AI in conversation. The capability set has matured. Australian tenants that have a Microsoft 365 Copilot rollout in flight or any policy on staff use of consumer ChatGPT, Claude or Gemini should have it on by now. Many do not.

Frontrow has run the setup against enough Australian tenants now to publish the working sequence. The product itself is well documented inside Microsoft Learn. What follows is the order of operations that lands cleanly in an AU mid-market or enterprise tenant, the licensing reality, the policies worth turning on first, and the prerequisites most teams discover the hard way.

What the AI Hub actually does

DSPM for AI gives the security and compliance team a single pane of glass over how generative AI is being used inside the tenant. It surfaces interactions with Microsoft 365 Copilot, Copilot Studio agents, Microsoft Copilot, Microsoft Security Copilot and a growing list of third-party AI applications that staff reach through the browser. It surfaces the sensitive-data exposure inside those interactions. It runs ready-to-deploy policies that detect risky AI use, label sensitive AI prompts and responses, and feed the findings into Insider Risk Management. It is the closest thing Microsoft has shipped to an enterprise control plane for generative AI.

The licensing reality

There are two practical entry points. The fuller capability set sits behind Microsoft 365 E5, the E5 Compliance add-on, or the standalone Microsoft Purview compliance licences. That tier unlocks auto-labelling at scale, Insider Risk Management for AI, advanced audit with extended retention, and the policy depth that makes the platform earn its keep at enterprise scale.

A lighter version of DSPM for AI is available to tenants on Microsoft 365 E3 with Microsoft 365 Copilot licences. That tier shows the AI activity, the exposure surface and the basic policy templates, which is enough to make the case for the upgrade. For most Australian mid-market tenants Frontrow recommends turning the lighter version on inside a Copilot pilot, then bringing the E5 Compliance add-on into scope before the rollout crosses 200 seats.

Prerequisites worth landing first

• Microsoft Purview Audit at the higher tier with at least 180 days of retention. DSPM for AI's analytics depend on it.
• A baseline sensitivity label taxonomy in Purview Information Protection, even if it is the four-label Frontrow default (Public, Internal, Confidential, Highly Confidential). The AI Hub gets meaningfully more useful once interactions can be classified against labels.
• SharePoint Restricted Search enabled on any tenant where the Copilot rollout is mid-flight. This is the scoping control that most organisations have not flipped on, and the AI Hub's exposure data will be noisier without it.
• Microsoft Defender for Cloud Apps connected and discovering shadow AI usage at the network edge. The third-party AI app surface inside DSPM for AI is much richer when this is in place.
• Insider Risk Management policies seeded with at least one base policy, so the AI Hub findings can flow into existing investigation workflows rather than sitting in a separate inbox.

The first three policies to enable

DSPM for AI ships with a library of one-click policy templates. Most tenants leave too many off because the operational discipline to handle the alerts is not yet in place. Frontrow's standing recommendation is to enable three first and add the rest as the team builds the muscle.

1. 1Detect sensitive interactions with AI. The base policy that flags prompts or responses containing high-sensitivity data types defined in Purview. This is the canary in the coal mine.
2. 2Detect risky AI usage. The companion policy that catches patterns associated with potential data exfiltration through AI tools. Tunes within the first two weeks.
3. 3Audit Microsoft 365 Copilot activity. The policy that ensures every Copilot interaction is captured for the audit trail required by APRA, OAIC and most procurement and insurance reviews this year.

What gets missed the first time through

• Admin units. The newer support for admin units in DSPM for AI lets large Australian tenants delegate scope by business unit or geography. Most setup runs Frontrow has been called into used a single global scope and then had to retrofit the structure later. Decide on the scoping model before turning policies on.
• Sensitivity labels in agents. Copilot Studio agents inherit the tenant's labels, but only after the labels are correctly applied to the underlying sources. Skipping the labels program means DSPM for AI shows AI usage, but cannot tell whether it is sensitive AI usage.
• The third-party AI app catalogue. The default catalogue is broad but not exhaustive. Tenants with specialised SaaS that has added an AI feature recently should request additional apps be added to the catalogue and check coverage monthly.
• Reporting cadence. The data is only useful if someone reads it. Frontrow's standing recommendation is a 30-minute weekly review for the first quarter, dropping to fortnightly once the policy noise has been tuned out.

Frontrow runs the AI Hub setup as a structured engagement against an Australian tenant over a fortnight, paired with the Copilot readiness work where the two overlap. Phone 1300 012 466 or send a note through the contact page if the rollout is in motion and the control plane has not yet caught up.